A stochastic pulse bit-stream with high accurate multiplication

Abstract

In order to scale down a circuit size for hardware realization of neural network systems, there are many researches on a conventional pulse model neuron, in which, a conventional numerical calculation is replaced by a stochastic calculation. In the conventional stochastic calculation, LFSR (linear feedback shift register) has been used as a random number generator. In general, a good randomness is required for an exact calculation result. However, since the random numbers which are generated by LFSR is lacking in the randomness, there is a problem which affects an accuracy of the conventional stochastic calculation which uses LFSR for generating the random numbers. In our work, in order to improve the accuracy of the conventional stochastic calculation, we propose to use GLFSR (generalized-LFSR) which generates the random numbers which are rich in randomness as the random number generator for the stochastic calculation. The accuracy of the stochastic calculation is also evaluated using a standard deviation value (SD) and a maximum error. As a result, the proposed stochastic calculation which using GLFSR as the random number generator shows a highest accuracy of the calculation result.